Quartet Based Gene Tree Imputation Using Deep Learning Improves Phylogenomic Analyses Despite Missing Data.

Area of Science:

• Phylogenomics
• Computational Biology
• Machine Learning

Background:

• Species tree estimation relies on phylogenomic methods using multiple genes.
• Incomplete gene trees, due to sampling or biological factors, can reduce phylogenetic inference accuracy.

Purpose of the Study:

• Introduce and address the challenge of imputing quartet distributions from incomplete gene trees.
• Develop a deep learning method to improve species tree accuracy.

Main Methods:

• Present Quartet based Gene tree Imputation using Deep Learning (QT-GILD), an unsupervised deep learning approach.
• Utilize natural language processing techniques to learn and complete quartet distributions from incomplete gene trees.
• QT-GILD requires no explicit modeling of missing data or gene tree heterogeneity.

Main Results:

• QT-GILD effectively imputes missing quartet data, significantly improving species tree accuracy on simulated and empirical datasets.
• The method successfully accounts for gene tree estimation errors.
• Demonstrates a dramatic improvement in species tree accuracy.

Conclusions:

• QT-GILD represents a significant advancement in species tree estimation from incomplete gene data.
• The technique offers a general-purpose solution for handling missing data in phylogenomics.
• Advances the state-of-the-art in phylogenomic data imputation and species tree reconstruction.